Monitoring system for adverse reaction of instant drug
By building a drug-symptom-user map and using graph neural networks and dynamic Bayesian networks, the problem of difficulty in monitoring and predicting adverse reactions to fast-hair drugs in real time in the existing technology is solved, and accurate prediction of adverse reactions to drug drugs and personalized safety suggestions are achieved.
Patent Information
- Application Number
- CN202510048613.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-13
- Publication Date
- 2025-05-13
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing drug adverse reaction monitoring systems are difficult to capture and predict fast-hair adverse reactions in real time, and lack effective systems to simulate the complex causal relationship between drugs and symptoms.
By constructing a drug-symptom-user map, using graph neural network model and dynamic Bayesian network, combining incremental learning algorithms, real-time analysis and prediction of drug adverse reactions, and providing personalized drug safety suggestions.
Real-time monitoring and accurate prediction of adverse reactions to fast-hair drugs is achieved, reducing the probability of adverse reactions, and improving drug safety and patient satisfaction.
Smart Images
Figure CN119993555A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of adverse drug reactions, in particular to a monitoring system for rapid-onset adverse drug reactions. Background Art
[0002] In the field of drug safety monitoring, with the development of medical technology and the continuous emergence of new drugs, the monitoring of adverse drug reactions has become a key link in protecting public health. Immediate adverse reactions refer to adverse reactions that occur within a short period of time after taking a drug. Such reactions often need to be quickly identified and handled to reduce the impact on the patient's health.
[0003] Although the existing monitoring system has played a certain role in monitoring adverse drug reactions, there are still some obvious defects. Traditional monitoring systems often rely on static data collection and analysis methods, lacking the ability to capture and respond to the dynamic changes between drug use and adverse reactions in real time, resulting in the inability to capture rapid adverse reactions in a timely manner; when dealing with the time dependency between drugs and symptoms, existing technologies often lack an effective system to simulate this complex causal relationship, limiting the in-depth exploration of the potential connection between drugs, symptoms and users. Summary of the invention
[0004] 1. Technical issues to be resolved
[0005] In view of the deficiencies in the prior art, the present invention provides a monitoring system for immediate adverse drug reactions, which solves the problem of how to achieve real-time monitoring and accurate prediction of immediate adverse drug reactions by building a model.
[0006] (II) Technical solution
[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A monitoring system for rapid-onset adverse drug reactions, comprising:
[0008] Collect user feedback on adverse drug reactions and obtain data on immediate adverse drug reactions after taking drugs, including the user's physiological reactions, symptoms and related historical drug records;
[0009] Construct a drug-symptom-user map. By processing the collected adverse reaction feedback, a multi-dimensional relationship map between drugs, symptoms and users is established. This map expresses the immediate adverse reaction symptoms of different users after taking specific drugs, and the impact of the interaction between different drugs on adverse reaction symptoms. The map is used to deeply analyze and mine the hidden associations between drugs and the complex relationship between symptom manifestations;
[0010] Based on the drug-symptom-user graph, a graph neural network model is used for data analysis. The constructed graph is analyzed using a graph neural network to explore the potential connections between drugs, symptoms, and users. In combination with individual differences and historical data of users, the interaction between drug combinations is further used to predict immediate adverse reactions and assess whether the drug combination causes adverse reactions, thereby warning of potential drug safety risks.
[0011] Provide drug safety recommendations based on the results of the graph neural network model. Based on the graph neural network analysis results, analyze drug usage plans for users, including safety tips for drug adjustment, dosage optimization, or combined use, to reduce the probability of rapid adverse reactions.
[0012] Collect users' feedback on adverse reactions after taking the medicine. By writing crawler scripts, use the HTTP protocol to actively crawl data from clinical data storage systems and online medical databases, including the hospital's electronic health record system and patient portal website; set up online questionnaires and feedback mechanisms to allow users to input their own physiological reactions and symptom manifestations. For example, users complete an online questionnaire that contains predefined questions, such as, "Do you feel uncomfortable after taking XX medicine?" and a drop-down menu listing possible symptoms. Users can select or enter the specific symptoms they have experienced; use text mining technology and natural language processing libraries such as NLTK or spaCy to perform semantic analysis and sentiment analysis on users' online feedback and clinical records; semantic analysis extracts key information such as drug names and symptom descriptions in the text through named entity recognition, while sentiment analysis uses sentiment classification algorithms such as VADER or TextBl ob, to identify the emotional tendencies in user feedback; for example, the system will analyze sentences in user feedback, some of which are "I felt very nauseous after taking the medicine", and then identify the nausea field as a symptom, while the sentiment analysis tool will identify the very field as a modifier that strengthens negative emotions; in the data cleaning and preprocessing stage, SQL queries and Python scripts are used to deduplicate data, correct errors, and standardize data; for example, for drug names, a pre-defined comparison table of drug names and standardized drug codes is used to convert the drug names entered by users into unified drug codes; for symptom descriptions, stemming and word form restoration are used to restore different word forms to basic forms for easy analysis and comparison.
[0013] The process of constructing the drug-symptom-user graph extracts key information from user feedback through entity recognition; for example, using pre-trained models such as BERT or SpaCy, the user feedback text is analyzed to identify entities such as drug names, symptom descriptions, and user information, and these entities are used as nodes in the graph; a dynamic Bayesian network is used to model the causal relationship between nodes, and the causal edge of each node in the graph represents a conditional probability, which is learned and updated based on the user's feedback data through the maximum likelihood estimation method; for example, if user feedback shows that headache symptoms often occur after taking a certain drug, then in the graph, the conditional probability of the causal edge between the drug node and the headache symptom node will increase accordingly; combined with the incremental learning algorithm The graph is dynamically adapted to new adverse reaction data. When new data is received, the incremental learning algorithm automatically updates the weights of causal edges by monitoring changes in the strength of associations between nodes without reconstructing the graph. For example, if new data show that a new symptom is more closely associated with a specific drug, the weight of the corresponding causal edge in the graph will increase. The accuracy of causal edges is enhanced by capturing the temporal dependency between drug use and symptom onset. This is achieved by introducing a time decay factor in the dynamic Bayesian network, which simulates the time delay effect of symptom onset. For example, if a symptom usually occurs within 24 hours after taking a medication, then in the dynamic Bayesian network, the causal edge between the symptom node and the drug node is weighted according to the time dependency.
[0014] The dynamic Bayesian network construction process first creates an initial Bayesian network model, whose nodes represent drug, symptom and user entities, and causal edges represent the probabilistic dependency relationship between these entities; the initial conditional probability table is set based on historical data, which comes from clinical trial results or previous patient feedback; natural language processing techniques, such as regular expression matching and machine learning classifiers, are used to extract entity information from user feedback and incorporate this information into the Bayesian network as nodes; for example, if the user feedback mentions that a headache occurs after taking aspirin, then aspirin and headache are added to the network as nodes, and the causal edge between them represents the occurrence of headache after taking aspirin. conditional probability; when new adverse reaction feedback is received, the confidence-based incremental update algorithm is used to update the Bayesian network; the algorithm monitors changes in confidence by comparing the consistency of new data with the existing conditional probability table; for example, if the new feedback shows that the association between aspirin and headache has increased, and this change is inconsistent with the existing probability table, the confidence will decrease; when the confidence drops to a preset threshold, the update mechanism will be triggered to update only the conditional probability table of the nodes directly related to the new data; the confidence is calculated by comparing the newly received adverse reaction feedback with the prediction results of the network's current conditional probability table. This comparison is achieved through a scoring function, which is calculated by the scoring function S = 1-|P 预测-P 实际 ∣, where S is the result of the scoring function, i.e., the confidence score; P 预测 is the probability of symptom occurrence predicted by the dynamic Bayesian network; P 实际 is the actual occurrence probability of the symptom in the new feedback data; for example, a threshold is set to 0.95, and only when the consistency between the new data and the existing model prediction is less than 95%, the confidence is considered to have dropped to the point where an update is required; for example, if according to the current conditional probability table, the network predicts that the probability of a headache after taking aspirin is 10%, but the new feedback shows that among the last 100 users who took aspirin, 16 reported a headache, which is much higher than the predicted 10%. In this case, the scoring function will calculate a confidence score, such as 0.94, which is lower than the preset threshold of 0.95, and thus triggers the update mechanism.
[0015] The update mechanism uses incremental learning and only adjusts nodes that are directly related to new feedback. For example, if new feedback shows that the occurrence of headaches is more closely associated with a specific drug, only the conditional probability table related to the drug and headaches will be updated. For the nodes that trigger the update, their conditional probabilities will be re-estimated to include the information of the new feedback while retaining the previously learned knowledge. This enables its dynamic Bayesian network to adapt to new data while maintaining stability and predictive accuracy, ensuring that the graph can reflect the latest relationship between drug use and adverse reactions in real time.
[0016] A graph neural network model is used to conduct in-depth analysis of the drug-symptom-user graph to achieve accurate prediction and analysis of adverse drug reactions. In this process, each node and causal edge in the graph is encoded as the input of the graph neural network model. The nodes represent drug, symptom or user entities, while the causal edges represent the relationship between these entities, such as the causal relationship between drug and symptom or the usage relationship between user and drug. The attribute information of each node and causal edge, such as the chemical structure of the drug, the descriptive text of the symptom and the user's medication history, is integrated into the graph neural network model to form a composite feature representation. This representation contains the information of the nodes and causal edges themselves, as well as the relationship information between them, providing contextual data for the model. In order to capture the local neighborhood information of the nodes and learn the complex dependencies between nodes, the graph neural network model uses the graph attention network structure to aggregate the neighbor node information of each node, thereby capturing the local structural characteristics of the node in the graph. For example, if a node represents the symptom of headache, the graph neural network model considers the information of the drug nodes and user nodes directly connected to the headache, as well as the neighbor node information of these nodes, to comprehensively evaluate the relationship between headache and specific drugs.
[0017] The graph neural network model dynamically adjusts the weights between nodes to more accurately mine the potential connections between drugs, symptoms and users; for example, if the model finds that the causal edge weight between a drug node and multiple symptom nodes increases, it indicates that there is a high correlation between the drug and these symptoms; combined with individual differences and historical data of users, the graph neural network model can predict the interactions between drug combinations and assess the risk of adverse reactions caused by drug combinations; for example, if a user node is connected to two drug nodes, the graph neural network model predicts the adverse reactions that may be caused by the combined use of these two drugs.
[0018] By simulating the drug's action process and combining the principles of pharmacodynamics and pharmacokinetic, a time series analysis of the occurrence of symptoms is performed. This analysis takes into account the absorption, distribution, metabolism, and excretion of drugs in the body, and how these processes affect the occurrence of symptoms. For example, if a user reports specific symptoms when the blood concentration of a drug reaches its peak in the body, the graph neural network model identifies this time dependency through time series analysis and accurately represents the causal relationship between the drug and the symptoms in the graph neural network model.
[0019] (III) Beneficial effects
[0020] The present invention provides a monitoring system for rapid-onset adverse drug reactions, which has the following beneficial effects:
[0021] 1. The present invention realizes the real-time update and dynamic adjustment of the drug-symptom-user map by combining a dynamic Bayesian network and an incremental learning algorithm; the map can respond quickly to new adverse reaction data without the need to reconstruct the map, thereby significantly improving the real-time and dynamic performance of the system.
[0022] 2. The present invention uses a graph neural network model to deeply analyze individual differences and historical data, providing each user with personalized drug safety recommendations, reducing the probability of immediate adverse reactions, improving treatment effects and patient satisfaction, while reducing medical costs and risks.
[0023] 3. The present invention constructs a drug-symptom-user graph and applies a graph neural network model to effectively mine the potential connections between drugs, symptoms and users, and evaluates the risk of adverse reactions caused by drug combinations. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0026] The system actively captures clinical data and information from online medical databases through a network interface, such as obtaining data from the hospital's electronic health record system and patient portal. At the same time, the system sets up an online questionnaire and feedback mechanism to allow users to input their own physiological reactions and symptoms. For example, users can use an online questionnaire to report whether they feel uncomfortable after taking a specific medication, and select or enter the specific symptoms they have experienced.
[0027] The system adopts text mining technology and uses natural language processing libraries such as NLTK or spaCy to perform semantic analysis and sentiment analysis on users' online feedback and clinical records. Semantic analysis extracts key information such as drug names and symptom descriptions in the text through named entity recognition, while sentiment analysis uses sentiment classification algorithms such as VADER or TextB l ob to identify emotional tendencies in user feedback. For example, the system will analyze the sentence in user feedback "I felt very nauseous after taking the medicine" and identify nausea as a symptom. At the same time, the sentiment analysis tool will identify "very" as a modifier that reinforces negative emotions.
[0028] During the data cleaning and preprocessing stage, the system uses SQL queries and Python scripts to perform data deduplication, error correction and data standardization; for drug names, the system uses a pre-defined drug name and standardized drug code comparison table to convert the drug names entered by users into unified drug codes; for symptom descriptions, the system uses stem extraction and word form restoration technology to restore different word forms to basic forms for easy analysis and comparison.
[0029] The process of constructing the drug-symptom-user graph extracts key information from user feedback through entity recognition; for example, using pre-trained models such as BERT or SpaCy, the user feedback text is analyzed to identify entities such as drug names, symptom descriptions, and user information, and these entities are used as nodes in the graph; a dynamic Bayesian network is used to model the causal relationship between nodes, where the causal edge of each node represents a conditional probability, which is learned and updated based on the user's feedback data through the maximum likelihood estimation method; for example, if user feedback shows that headache symptoms often occur after taking a certain drug, then in the graph, the conditional probability of the causal edge between the drug node and the headache symptom node will increase accordingly.
[0030] In combination with an incremental learning algorithm, the graph is made to dynamically adapt to new adverse reaction data; when new data is received, the incremental learning algorithm automatically updates the weights of causal edges by monitoring changes in the strength of associations between nodes without having to reconstruct the graph; for example, if new data show that a new symptom has an increased association with a specific drug, then the corresponding causal edge weight in the graph will increase; the accuracy of causal edges is enhanced by capturing the temporal dependency between drug use and symptom onset, which is achieved by introducing a time decay factor in the dynamic Bayesian network to simulate the time delay effect of symptom onset; for example, if a symptom usually occurs within 24 hours after taking a medication, then in the dynamic Bayesian network, the causal edge between the symptom node and the drug node is weighted according to the time dependency.
[0031] The dynamic Bayesian network construction process first creates an initial Bayesian network model, whose nodes represent drug, symptom and user entities, and causal edges represent the probabilistic dependency relationship between these entities; the initial conditional probability table is set based on historical data, which comes from clinical trial results or previous patient feedback; natural language processing techniques, such as regular expression matching and machine learning classifiers, are used to extract entity information from user feedback and incorporate this information into the Bayesian network as nodes; for example, if user feedback mentions that headache occurs after taking aspirin, then aspirin and headache are added to the network as nodes, and the causal edge between them represents the conditional probability of headache after taking aspirin.
[0032] When new adverse reaction feedback is received, a confidence-based incremental update algorithm is used to update the Bayesian network; the algorithm monitors changes in confidence by comparing the consistency of new data with the existing conditional probability table; for example, if the new feedback shows that the association between aspirin and headache has increased, and this change is inconsistent with the existing probability table, the confidence will decrease; when the confidence drops to a preset threshold, the update mechanism will be triggered to update only the conditional probability table of the nodes directly related to the new data; the confidence is calculated by comparing the newly received adverse reaction feedback with the prediction results of the network's current conditional probability table. This comparison is achieved through a scoring function. The scoring function is custom or based on existing statistical methods; for example, a threshold of 0.95 is set, and only when the consistency between the new data and the existing model prediction is less than 95%, the confidence is considered to have dropped to the point where an update is required; for example, if according to the current conditional probability table, the network predicts that the probability of a headache after taking aspirin is 10%, but new feedback shows that among the last 100 users who took aspirin, 16 reported headaches, which is much higher than the predicted 10%. In this case, the scoring function will calculate a confidence score of, for example, 0.94, which is lower than the preset threshold of 0.95, thus triggering the update mechanism.
[0033] The update mechanism uses incremental learning and only adjusts nodes that are directly related to new feedback. For example, if new feedback shows that the occurrence of headaches is more closely associated with a specific drug, only the conditional probability table related to the drug and headaches will be updated. For the nodes that trigger the update, their conditional probabilities will be re-estimated to include the information of the new feedback while retaining the previously learned knowledge. This enables its dynamic Bayesian network to adapt to new data while maintaining stability and predictive accuracy, ensuring that the graph can reflect the latest relationship between drug use and adverse reactions in real time.
[0034] A graph neural network model is used to conduct in-depth analysis of the drug-symptom-user graph to achieve accurate prediction and analysis of adverse drug reactions. In this process, each node and causal edge in the graph is encoded as the input of the graph neural network model. The nodes represent drug, symptom or user entities, while the causal edges represent the relationship between these entities, such as the causal relationship between drug and symptom or the usage relationship between user and drug. The attribute information of each node and causal edge, such as the chemical structure of the drug, the descriptive text of the symptom and the user's medication history, is integrated into the graph neural network model to form a composite feature representation. This representation contains the information of the nodes and causal edges themselves, as well as the relationship information between them, providing contextual data for the model. In order to capture the local neighborhood information of the nodes and learn the complex dependencies between nodes, the graph neural network model uses the graph attention network structure to aggregate the neighbor node information of each node, thereby capturing the local structural characteristics of the node in the graph. For example, if a node represents the symptom of headache, the graph neural network model considers the information of the drug nodes and user nodes directly connected to the headache, as well as the neighbor node information of these nodes, to comprehensively evaluate the relationship between headache and specific drugs.
[0035] The graph neural network model dynamically adjusts the weights between nodes to more accurately mine the potential connections between drugs, symptoms and users; for example, if the model finds that the causal edge weight between a drug node and multiple symptom nodes increases, it indicates that there is a high correlation between the drug and these symptoms; combined with individual differences and historical data of users, the graph neural network model can predict the interactions between drug combinations and assess the risk of adverse reactions caused by drug combinations; for example, if a user node is connected to two drug nodes, the graph neural network model predicts the adverse reactions that may be caused by the combined use of these two drugs.
[0036] By simulating the drug's action process and combining the principles of pharmacodynamics and pharmacokinetic, a time series analysis of the occurrence of symptoms is performed. This analysis takes into account the absorption, distribution, metabolism, and excretion of drugs in the body, and how these processes affect the occurrence of symptoms. For example, if a user reports specific symptoms when the blood concentration of a drug reaches its peak in the body, the graph neural network model identifies this time dependency through time series analysis and accurately represents the causal relationship between the drug and the symptoms in the graph neural network model.
[0037] Suppose a patient reports that he or she has experienced symptoms of dizziness and nausea after taking a new antihypertensive drug; these feedbacks are collected through an online questionnaire, and text mining is performed using NLTK to identify dizziness and nausea as symptom entities, and antihypertensive drugs as drug entities; this information is integrated into the drug-symptom-user graph and triggers the update mechanism of the dynamic Bayesian network; the system evaluates the consistency of the new feedback with the existing conditional probability table and calculates the confidence score; if the score is lower than the preset threshold, the system will update the conditional probability table of the relevant node to reflect the new adverse reaction data; the graph neural network model further analyzes this data, mines the potential connection between drugs, symptoms and users, and predicts the risks that may be caused when the antihypertensive drug is used in combination with other drugs; finally, the system provides safety advice to doctors and patients to reduce the probability of adverse reactions; in this way, the monitoring system of the present invention can monitor and predict rapid-onset adverse drug reactions in real time, providing strong technical support for drug safety supervision.
[0038] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A monitoring system for rapid adverse drug reactions, characterized in that: include: Collect user feedback on adverse drug reactions and obtain data on immediate adverse drug reactions after taking drugs, including the user's physiological reactions, symptoms and related historical drug records; Construct a drug-symptom-user map. By processing the collected adverse reaction feedback, a multi-dimensional relationship map between drugs, symptoms and users is established. This map expresses the immediate adverse reaction symptoms of different users after taking specific drugs, and the impact of the interaction between different drugs on adverse reaction symptoms. The map is used to deeply analyze and mine the hidden associations between drugs and the complex relationship between symptom manifestations; Based on the drug-symptom-user graph, a graph neural network model is used for data analysis. The constructed graph is analyzed using a graph neural network to explore the potential connections between drugs, symptoms, and users. In combination with individual differences and historical data of users, the interaction between drug combinations is further used to predict immediate adverse reactions and assess whether the drug combination causes adverse reactions, thereby warning of potential drug safety risks. Provide drug safety recommendations based on the results of the graph neural network model. Based on the graph neural network analysis results, analyze drug usage plans for users, including safety tips for drug adjustments, dosage optimization, or combined use, to reduce the probability of rapid adverse reactions.
2. A monitoring system for rapid adverse drug reactions according to claim 1, characterized in that: The collection of user feedback on adverse drug reactions actively captures clinical data and information in online medical databases through a network interface, and sets up an online questionnaire and feedback mechanism to collect users' physiological reactions and symptom manifestations; Through text mining, semantic analysis and sentiment analysis are performed on users' online feedback and clinical records to extract key information and identify potential adverse reactions; the collected data are cleaned and preprocessed to ensure the quality of the collected data.
3. The system for monitoring rapid adverse drug reactions according to claim 1, characterized in that: The drug-symptom-user graph is constructed by combining a dynamic Bayesian network and an incremental learning algorithm; drug, symptom and user entities are extracted from user feedback through natural language processing technology, and these entities are used as nodes of the graph; a dynamic Bayesian network is used to model the causal relationship between nodes, wherein the causal edge of each node represents a conditional probability, and the probability is learned and updated based on user feedback data; the incremental learning algorithm is combined to enable the graph to dynamically adjust the weight of the causal edge when new adverse reaction data is received without reconstructing the graph, and the weight of the causal edge is automatically updated by monitoring the changes in the strength of association between nodes, so as to reflect the latest relationship between drug use and adverse reactions in real time; and the accuracy of the causal edge is further enhanced by capturing the time dependency between drug use and symptom onset.
4. A monitoring system for rapid-onset adverse drug reactions according to claim 3, characterized in that: The dynamic Bayesian network constructs an initial Bayesian network, in which nodes represent drugs, symptoms and user entities, and causal edges represent the probabilistic dependency relationship between entities; the initial conditional probability table is set based on historical data, entity information is extracted from user feedback, and the entity information is incorporated into the dynamic Bayesian network as nodes; when new adverse reaction feedback is received, an incremental update algorithm based on confidence is adopted, by monitoring the consistency of the new data with the existing conditional probability table, and the conditional probability table of the relevant node is updated only when the confidence drops to a preset threshold; local updates only affect nodes directly related to the new data to avoid frequent global retraining; the weights of causal edges are adjusted based on time decay to simulate the time dependency between drug use and symptom appearance; The weights of causal edges are adjusted over time based on the delayed effects of symptom onset.
5. A monitoring system for rapid-onset adverse drug reactions according to claim 4, characterized in that: The incremental update algorithm focuses on dynamically updating the conditional probability table of nodes in the Bayesian network. Based on the consistency evaluation of the newly received adverse reaction feedback data and the existing data, an initial threshold is set to determine when the map needs to be updated. For each new feedback data, a confidence score is calculated based on the degree of match between the new feedback data and the dynamic Bayesian network prediction; a scoring function is used to compare the symptoms in the new feedback data with the symptoms predicted by the dynamic Bayesian network based on the current conditional probability table. If the symptoms in the new feedback are within the confidence interval predicted by the dynamic Bayesian network, the new data is consistent with the dynamic Bayesian network and does not need to be updated; if not within the confidence interval, the confidence score is lower than the preset threshold, triggering the update mechanism; The update mechanism uses incremental learning, and only adjusts the nodes directly related to the new feedback data. For the nodes that trigger the update, their conditional probabilities are re-estimated to include the information of the new feedback data while retaining the previously learned knowledge.
6. A monitoring system for rapid adverse drug reactions according to claim 1, characterized in that: A graph neural network model is used to conduct an in-depth analysis of the drug-symptom-user graph. Each node and causal edge in the graph is encoded as the input of the graph neural network, combining the attribute information of the node with the relationship information of the causal edge to form a composite feature representation of each node and causal edge; capturing the local neighborhood information of the node and learning the complex dependencies between nodes; enabling the graph neural network to dynamically adjust the weights between nodes, thereby more accurately mining the potential connections between drugs, symptoms and users.
7. The system for monitoring rapid adverse drug reactions according to claim 1, characterized in that: The graph neural network model is used in combination with individual differences of users and historical data to predict the interactions between drug combinations and assess the risk of adverse reactions caused by drug combinations. By simulating the action process of drugs and combining the principles of pharmacodynamics and pharmacokinetic, a time series analysis of the occurrence of symptoms is performed, thereby accurately representing the causal relationship between drugs and symptoms in the graph neural network.